D&D General Can ChatGPT create a Campaign setting?

[LuisCarlos17f]

Interesting, but I warn all created by AI can't be protected by copyright.

 

[Ruin Explorer]

This is weirdly reassuring because it's so extremely uncreative/cliche and even the "fantasy food" is just desperately anti-immersive because it's all so extreme/lazy, like a imagine a RL restaurant where every single item was either themed or actually based on predatory megafauna. I'm sure just such a place exists somewhere, but it's going to be a ultra-tacky theme restaurant visited solely by tourists and the most awful kind of "bro".

 

[Jer]

This is weirdly reassuring because it's so extremely uncreative/cliche

Literally that's what you're going to get from an AI using current methods. In large data models the more examples you have of something occurring in your training set the more likely it will occur in the output. This will trend towards the system proposing things that are similar to but not exactly like things it's seen in its training set.

If you remember Alphabot from Futurama - the network executive bot that was programmed to like things that it had seen before - then you have the gist of what these models can ultimately be expected to produce. A lot of things that are like things they've seen before but just different enough to look somewhat novel.

(You also have to be careful - these models can potentially also reproduce EXACTLY what was in their training set given the right prompts, because they are to use a flawed analogy somewhat "memorizing" what they've seen before as they construct their models. So you can potentially unknowingly violate copyright if the model was trained on something under copyright and you accidentally reproduce it and publish it.)

 

[Grendel_Khan]

Truly though, what is the point of seeking out AI-generated setting content when there are more published settings than you could ever play in one lifetime? This is what I really don't get about most "creative" uses of ChatGPT—we don't have a scarcity problem with human-generated text.

Gross as I think AI-generated art is, that I can at least somewhat understand, because it's harder to affordably obtain quality artwork for your own use. But stuff like bog-standard generic fantasy settings. Just head to DriveThru, close your eyes, click at random, and you'll find something cheap and likely much more thoughtful than the slop that a large language model regurgitates.

 

[Reynard]

Truly though, what is the point of seeking out AI-generated setting content when there are more published settings than you could ever play in one lifetime? This is what I really don't get about most "creative" uses of ChatGPT—we don't have a scarcity problem with human-generated text.

Gross as I think AI-generated art is, that I can at least somewhat understand, because it's harder to affordably obtain quality artwork for your own use. But stuff like bog-standard generic fantasy settings. Just head to DriveThru, close your eyes, click at random, and you'll find something cheap and likely much more thoughtful than the slop that a large language model regurgitates.

But it isn't a random desire. More analogous would be a completely indexed set of every published setting ever that someone can search with phrasing. "Give me sky pirates versus dragons in an aetherpunk world that's a moon to an atomic age planet." If that setting exists, the search engine could point you to it.

 

[Grendel_Khan]

But it isn't a random desire. More analogous would be a completely indexed set of every published setting ever that someone can search with phrasing. "Give me sky pirates versus dragons in an aetherpunk world that's a moon to an atomic age planet." If that setting exists, the search engine could point you to it.

That definitely sounds useful. But have you used ChatGPT? That's not how it or Bard or any other large language model works. One of the biggest problems with ChatGPT is that it doesn't—arguably can't—give you sources or point you to what it's ingested.

And even if you interpret its output as a kind of non-sourced search result, that's also not how those models work, because it's going to throw everything seemingly relevant into a blender of language correlations, mixing some human's possibly well-considered sky-pirate aetherpunk setting with random bits and bobs, none of it based on any sort of reason or cognition, just what words seem like they might go together in convincing ways.

 

[Andvari]

Interestingly, I asked it to create a post-apocalyptic viking setting and it named the setting "Risengard". I asked how it came up with the name and got the following response.

The name "Risengard" was created as a nod to Norse mythology and the concept of the new world after Ragnarok. In Norse mythology, "Rise" means "dawn" or "daybreak", symbolizing the start of a new era. "Gard" is a reference to "Midgard", one of the Nine Realms in Norse cosmology, meaning "middle yard" or "middle earth". So, "Risengard" means "dawn of the middle earth" or "new middle earth". I thought it would be a fitting name for a world that has risen from the ashes of Ragnarok, marking the start of a new era.

 

[Clint_L]

That definitely sounds useful. But have you used ChatGPT? That's not how it or Bard or any other large language model works. One of the biggest problems with ChatGPT is that it doesn't—arguably can't—give you sources or point you to what it's ingested.

And even if you interpret its output as a kind of non-sourced search result, that's also not how those models work, because it's going to throw everything seemingly relevant into a blender of language correlations, mixing some human's possibly well-considered sky-pirate aetherpunk setting with random bits and bobs, none of it based on any sort of reason or cognition, just what words seem like they might go together in convincing ways.

It absolutely can point you to sources; I routinely use it for that purpose.

It is a creative tool. You have to think of it as a personal assistant, and you as the director. You supply the ideas, and get it to rapidly iterate on them. You could 100% use it to generate an extremely detailed and original setting, and I am sure lots of people already are.

Edit: also, I think it is very likely that you can copyright material created with it, depending on the degree of human involvement, though I am sure this area of law is about to see a ton of litigation.

 

[Burt Baccara]

If you asked me this a month ago, I would have said yes. Lately, Chat-GPT seems to be forgetting the thread of a conversation.

Another aspect to remember, if the plan is to publish this work, the work created by an AI can't be copyrighted, though that may change due to a comic book called Zarya Of The Dawn by Kris Kashtanova.

 

[Reynard]

That definitely sounds useful. But have you used ChatGPT? That's not how it or Bard or any other large language model works. One of the biggest problems with ChatGPT is that it doesn't—arguably can't—give you sources or point you to what it's ingested.

And even if you interpret its output as a kind of non-sourced search result, that's also not how those models work, because it's going to throw everything seemingly relevant into a blender of language correlations, mixing some human's possibly well-considered sky-pirate aetherpunk setting with random bits and bobs, none of it based on any sort of reason or cognition, just what words seem like they might go together in convincing ways.

I wasn't saying that's how ChatGPT worked. I was saying that people using ChatGPT to create a setting would not be served by randomly picking a setting from DriveThru, which was the claim you made. People want something specific. That's why they are using an AI to generate something instead of just googling for it.